CN106980854A - Number-plate number recognition methods, device, storage medium and processor - Google Patents
Number-plate number recognition methods, device, storage medium and processor Download PDFInfo
- Publication number
- CN106980854A CN106980854A CN201710198950.3A CN201710198950A CN106980854A CN 106980854 A CN106980854 A CN 106980854A CN 201710198950 A CN201710198950 A CN 201710198950A CN 106980854 A CN106980854 A CN 106980854A
- Authority
- CN
- China
- Prior art keywords
- convolutional neural
- neural networks
- networks model
- area image
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/625—License plates
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Character Discrimination (AREA)
Abstract
The invention discloses a kind of number-plate number recognition methods, device, storage medium and processor.Wherein, this method includes：Obtain by camera acquisition to history vehicle image in the first license plate area image positional information and number information；The first default convolutional neural networks model is trained according to the first license plate area image and positional information；Handle pre-set image model, the second default convolutional neural networks model and reach that the first default convolutional neural networks model of convergence state is cascaded, so as to obtain the 3rd default convolutional neural networks model；The 3rd default convolutional neural networks model is trained according to the second license plate area image and number information；The Current vehicle image collected is identified according to the reach convergence state the 3rd default convolutional neural networks model, the recognition result of the number-plate number in Current vehicle image is obtained.The present invention solves the recognition accuracy and less efficient technical problem that Car license recognition is present in the prior art.
Description
Technical field
The present invention relates to field of traffic, in particular to a kind of number-plate number recognition methods, device, storage medium and
Processor.
Background technology
With the fast development and extensive use of electronics and information industry, wisdom traffic has become the mark of Modern Traffic system
Will.Wherein, Car license recognition is the important component of wisdom traffic, by being adopted to equipment such as traffic camera, drive recorders
The image information or video information collected is handled, the car plate and number that identify vehicle in image that can automatically, intelligent
Information.Car license recognition can be efficiently applied to automatic traffic illegal activities identification, suspect vehicle target following, vehicle retrieval, car
The scenes such as toll administration, wisdom parking lot, district vehicles management.
At present, existing license plate recognition technology is roughly divided into following two：Car license recognition side based on traditional images processing
Method and the licence plate recognition method based on deep learning.
Method one, the licence plate recognition method based on traditional images processing can realize that car plate is determined using traditional images algorithm
Position, Character segmentation and character recognition.Conventional traditional algorithm of locating license plate of vehicle includes being based on car plate textural characteristics, color characteristic, side
Edge feature, genetic algorithm etc.；Conventional conventional characters partitioning algorithm includes method based on projection etc.；Conventional conventional characters are known
Other algorithm includes method based on character stroke etc..Due to the unicity of vehicle license plate characteristic, traditional car plate based on image procossing
Identification can also obtain good effect under qualifications, but its parameter is more, and regulation optimization is sufficiently complex manually, and to image
Quality requirement is very high, under different application scenarios (including extremely complicated scene or special screne) reliability and applicability compared with
Difference, recognition accuracy and recognition efficiency are relatively low.
Method two, the licence plate recognition method based on deep learning can be by building the convolutional neural networks of deep layer, to defeated
The license plate image entered carries out feature extraction and abstract analysis, realizes the automatic detection identification of the number-plate number.For example, patent publication No.
It is for CN104298976A, patent name《Detection method of license plate based on convolutional neural networks》Patent document in using being based on
The Adaboost of Haar features carries out license plate area rough detection, and is accurately detected with reference to convolutional neural networks CNN, and then adopts
Character segmentation is carried out with conventional threshold values split plot design, the single character picture after most splitting at last is input to follow-up convolutional neural networks
Model is identified.But, detection and segmentation performance in this kind of method are limited by conventional method, introduce secondary mistake
Difference, so as to reduce the accuracy of testing result.
Therefore, there is reliability in the recognition methods of the both the above number-plate number and applicability is poor, is easily introduced second order error, knows
The other degree of accuracy and the relatively low defect of recognition efficiency, to sum up, there is recognition accuracy and knowledge in licence plate recognition method of the prior art
Not less efficient technical problem.
For it is above-mentioned the problem of, effective solution is not yet proposed at present.
The content of the invention
The embodiments of the invention provide a kind of number-plate number recognition methods, device, storage medium and processor, at least to solve
Recognition accuracy and less efficient technical problem that certainly Car license recognition is present in the prior art.
One side according to embodiments of the present invention includes there is provided a kind of number-plate number recognition methods, this method：Obtain
By camera acquisition to history vehicle image in the first license plate area image positional information and number information；According to above-mentioned
First license plate area image and above-mentioned positional information are trained to the first default convolutional neural networks model, until above-mentioned first
Default convolutional neural networks model reaches convergence state, wherein, the above-mentioned first default convolutional neural networks model is used for above-mentioned
Detected car plate position in first license plate area image；To pre-set image processing model, the second default convolutional neural networks
Model is cascaded with the above-mentioned first default convolutional neural networks model for reaching above-mentioned convergence state, is preset so as to obtain the 3rd
Convolutional neural networks model, wherein, above-mentioned pre-set image processing model is used to align to above-mentioned first license plate area image
Processing is so as to obtain the second license plate area image, and the above-mentioned second default convolutional neural networks model is used for above-mentioned second car plate area
The number-plate number in area image carries out depth recognition；According to above-mentioned second license plate area image and above-mentioned number information to above-mentioned
Three default convolutional neural networks models are trained, until the above-mentioned 3rd default convolutional neural networks model reaches convergence state；
The Current vehicle image collected is carried out according to the reach above-mentioned convergence state the above-mentioned 3rd default convolutional neural networks model
Identification, obtains the recognition result of the above-mentioned number-plate number in above-mentioned Current vehicle image.
Further, convolutional Neural net is being preset to first according to above-mentioned first license plate area image and above-mentioned positional information
Before network model is trained, the above method also includes：To the network parameter in the above-mentioned first default convolutional neural networks model
Carry out initialization process.
Further, it is above-mentioned that convolutional Neural is preset to first according to above-mentioned first license plate area image and above-mentioned positional information
Network model be trained including：Above-mentioned car plate position in above-mentioned first license plate area image and the above-mentioned number-plate number are distinguished
It is marked, obtains car plate location tags and number-plate number label；According to above-mentioned car plate location tags and above-mentioned number-plate number mark
Label set up license board information database, wherein, multiple data samples are included in above-mentioned license board information database；According to above-mentioned data sample
Convolutional neural networks model is preset in this training above-mentioned first.
Further, refreshing to the above-mentioned 3rd default convolution according to above-mentioned second license plate area image and above-mentioned number information
Before being trained through network model, the above method also includes：Model is handled according to above-mentioned pre-set image default to above-mentioned first
The above-mentioned first license plate area image of convolutional neural networks model output carries out two dimensional affine conversion process, obtains above-mentioned second car
Board area image.
Another aspect according to embodiments of the present invention, additionally provides a kind of number-plate number identifying device, and the device includes：Obtain
Take unit, for obtain by camera acquisition to history vehicle image in the first license plate area image positional information and number
Code information；First processing units, for presetting convolution to first according to above-mentioned first license plate area image and above-mentioned positional information
Neural network model is trained, until the above-mentioned first default convolutional neural networks model reaches convergence state, wherein, above-mentioned the
One default convolutional neural networks model is used to detect the car plate position in above-mentioned first license plate area image；Second processing
Unit, for handling pre-set image model, the second default convolutional neural networks model and reaching the above-mentioned of above-mentioned convergence state
First default convolutional neural networks model is cascaded, so that the 3rd default convolutional neural networks model is obtained, wherein, it is above-mentioned pre-
If image processing model is used to above-mentioned first license plate area image is carried out registration process to obtain the second license plate area image,
Above-mentioned second default convolutional neural networks model is used to carry out depth knowledge to the number-plate number in above-mentioned second license plate area image
Not；3rd processing unit, for presetting convolution to the above-mentioned 3rd according to above-mentioned second license plate area image and above-mentioned number information
Neural network model is trained, until the above-mentioned 3rd default convolutional neural networks model reaches convergence state；Recognition unit, is used
The Current vehicle image collected is entered according to the reach above-mentioned convergence state the above-mentioned 3rd default convolutional neural networks model
Row identification, obtains the recognition result of the above-mentioned number-plate number in above-mentioned Current vehicle image.
Further, said apparatus also includes：Fourth processing unit, for the above-mentioned first default convolutional neural networks mould
Network parameter in type carries out initialization process.
Further, above-mentioned first processing units include：Subelement is marked, in above-mentioned first license plate area image
Above-mentioned car plate position and the above-mentioned number-plate number be marked respectively, obtain car plate location tags and number-plate number label；Create
Subelement, for setting up license board information database according to above-mentioned car plate location tags and above-mentioned number-plate number label, wherein, it is above-mentioned
Multiple data samples are included in license board information database；First processing subelement, for training above-mentioned according to above-mentioned data sample
First default convolutional neural networks model.
Further, said apparatus also includes：5th processing unit, for according to above-mentioned pre-set image handle model to
The above-mentioned first license plate area image for stating the first default convolutional neural networks model output carries out two dimensional affine conversion process, obtains
Above-mentioned second license plate area image.
Another aspect according to embodiments of the present invention, additionally provides a kind of storage medium, it is characterised in that above-mentioned storage is situated between
Matter includes the program of storage, wherein, equipment where above-mentioned storage medium is controlled when said procedure is run performs above-mentioned car plate
Number identification method.
Another aspect according to embodiments of the present invention, additionally provides processor, it is characterised in that above-mentioned processor is used to transport
Line program, wherein, said procedure performs the above-mentioned number-plate number recognition methods when running.
In embodiments of the present invention, using obtain by camera acquisition to history vehicle image in the first license plate area
The positional information of image and the mode of number information, and then according to the first license plate area image and positional information to the first default volume
Product neural network model is trained, until the first default convolutional neural networks model reaches convergence state, and then by pre-
If image processing model, the second default convolutional neural networks model and the first default convolutional neural networks mould for reaching convergence state
Type carries out cascade so as to obtain the 3rd default convolutional neural networks model, and then according to the second license plate area image and number information
3rd default convolutional neural networks model is trained up to the 3rd default convolutional neural networks model reaches convergence state, reached
The Current vehicle image collected is identified to convolutional neural networks model is preset according to the reach convergence state the 3rd
So that the purpose of the recognition result of the number-plate number in Current vehicle image is obtained, it is achieved thereby that lifting number-plate number identification
The degree of accuracy and recognition efficiency, avoid the extraneous unfavorable factor such as environment, illumination, fuzzy interference technique effect, and then solve
Car license recognition is present in the prior art recognition accuracy and less efficient technical problem.
Brief description of the drawings
Accompanying drawing described herein is used for providing a further understanding of the present invention, constitutes the part of the application, this hair
Bright schematic description and description is used to explain the present invention, does not constitute inappropriate limitation of the present invention.In the accompanying drawings：
Fig. 1 is a kind of schematic flow sheet of optional number-plate number recognition methods according to embodiments of the present invention；
Fig. 2 is the schematic flow sheet of another optional number-plate number recognition methods according to embodiments of the present invention；
Fig. 3 is the schematic flow sheet of another optional number-plate number recognition methods according to embodiments of the present invention；
Fig. 4 is the structural representation that according to embodiments of the present invention a kind of optional first presets convolutional neural networks model
Figure；
Fig. 5 is a kind of structural representation of optional number-plate number identifying device according to embodiments of the present invention.
Embodiment
In order that those skilled in the art more fully understand the present invention program, below in conjunction with the embodiment of the present invention
Accompanying drawing, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is only
The embodiment of a part of the invention, rather than whole embodiments.Based on the embodiment in the present invention, ordinary skill people
The every other embodiment that member is obtained under the premise of creative work is not made, should all belong to the model that the present invention is protected
Enclose.
It should be noted that term " first " in description and claims of this specification and above-mentioned accompanying drawing, "
Two " etc. be for distinguishing similar object, without for describing specific order or precedence.It should be appreciated that so using
Data can exchange in the appropriate case, so as to embodiments of the invention described herein can with except illustrating herein or
Order beyond those of description is implemented.In addition, term " comprising " and " having " and their any deformation, it is intended that cover
Lid is non-exclusive to be included, for example, the process, method, system, product or the equipment that contain series of steps or unit are not necessarily limited to
Those steps or unit clearly listed, but may include not list clearly or for these processes, method, product
Or the intrinsic other steps of equipment or unit.
Embodiment 1
According to embodiments of the present invention there is provided a kind of embodiment of number-plate number recognition methods, it is necessary to illustrate, attached
The step of flow of figure is illustrated can perform in the computer system of such as one group computer executable instructions, though also,
So logical order is shown in flow charts, but in some cases, can be shown to be performed different from order herein
Or the step of description.
Fig. 1 is a kind of schematic flow sheet of optional number-plate number recognition methods according to embodiments of the present invention, such as Fig. 1 institutes
Show, this method comprises the following steps：
Step S102, obtain by camera acquisition to history vehicle image in the first license plate area image position believe
Breath and number information；
Step S104, is carried out according to the first license plate area image and positional information to the first default convolutional neural networks model
Training, until the first default convolutional neural networks model reaches convergence state, wherein, the first default convolutional neural networks model is used
Detect car plate position in the first license plate area image；
Step S106, handles pre-set image model, the second default convolutional neural networks model and reaches convergence state
First default convolutional neural networks model is cascaded, so that the 3rd default convolutional neural networks model is obtained, wherein, preset figure
As processing model is used for the first license plate area image progress registration process so as to obtain the second license plate area image, second presets
Convolutional neural networks model is used to carry out depth recognition to the number-plate number in the second license plate area image；
Step S108, is carried out according to the second license plate area image and number information to the 3rd default convolutional neural networks model
Training, until the 3rd default convolutional neural networks model reaches convergence state；
Step S110, according to the reach convergence state the 3rd default convolutional neural networks model to the Current vehicle that collects
Image is identified, and obtains the recognition result of the number-plate number in Current vehicle image.
In embodiments of the present invention, using obtain by camera acquisition to history vehicle image in the first license plate area
The positional information of image and the mode of number information, and then according to the first license plate area image and positional information to the first default volume
Product neural network model is trained, until the first default convolutional neural networks model reaches convergence state, and then by pre-
If image processing model, the second default convolutional neural networks model and the first default convolutional neural networks mould for reaching convergence state
Type carries out cascade so as to obtain the 3rd default convolutional neural networks model, and then according to the second license plate area image and number information
3rd default convolutional neural networks model is trained up to the 3rd default convolutional neural networks model reaches convergence state, reached
The Current vehicle image collected is identified to convolutional neural networks model is preset according to the reach convergence state the 3rd
So that the purpose of the recognition result of the number-plate number in Current vehicle image is obtained, it is achieved thereby that lifting number-plate number identification
The degree of accuracy and recognition efficiency, avoid the extraneous unfavorable factor such as environment, illumination, fuzzy interference technique effect, and then solve
Car license recognition is present in the prior art recognition accuracy and less efficient technical problem.
Alternatively, the shortcoming based on prior art is not enough, and present applicant proposes a kind of end-to-end car based on deep learning
Board number identification method, by building the convolutional neural networks model of car plate detection and Number Reorganization, is prevented effectively from environment, light
According to, the interference brought such as fuzzy, it is ensured that the precision of identification.Meanwhile, the application need not carry out Character segmentation, and draw in recognition result
Question mark character is entered, the Number Reorganization under the conditions of solving the problems, such as part car plate, block etc. has Shandong in a variety of application environments
Rod.
Alternatively, during step S102 is performed, it can be obtained by camera, Internet resources and include vehicle and car
First license plate area image of board, the first license plate area image can be referred to as image sample data again.And then, to image sample
Notebook data carries out postsearch screening, removes picture quality and the undesirable set of view data of image content, can obtain M opening and closing lattice
Image sample data, finally, can be to garbled M image sample data addition car plate positions and number-plate number label.
Alternatively, car plate location tags include the position of all car plates in image, and each position is made up of two coordinates
Four points, respectively car plate edge frame the upper left corner and the lower right corner；Number-plate number label includes number of all car plates in image
Code, 7 characters that each number is made up of Chinese character, numeral, letter and question mark, wherein question mark is represented due to obscuring, blocking
The character that problem is beyond recognition.
Alternatively, chinese character can include 31 Chinese characters：Capital, Shanghai, Tianjin, Chongqing, Ji, Shanxi, illiteracy, the Liao Dynasty, Ji, black, Soviet Union, Zhejiang,
It is Anhui, Fujian, Jiangxi, Shandong, Henan, Hubei Province, Hunan, Guangdong, osmanthus, fine jade, river, expensive, cloud, Tibetan, Shan, sweet, blue or green, peaceful, new；Numerical character can include 10
Numeral：0、1、2、3、4、5、6、7、8、9；Alphabetic character can include 24 English alphabets：A、B、C、D、E、F、G、H、J、K、L、
M、N、P、Q、R、S、T、U、V、W、X、Y、Z。
Alternatively, the first default convolutional neural networks model can be pre-established.Specifically, deep learning can be used
Technology builds multilayer convolutional neural networks, including input data layer, Nc convolutional layer, Np pond layer, a Nf full connections
Layer.Wherein, every layer of convolutional layer includes some convolution kernels, and i-th of convolutional layer includes Ki convolution kernel, and the size of convolution kernel is Ks_
I*Ks_i, step-length is Kb_i.Every layer of pond layer is using maximum pond method, and Chi Huahe size is Kps_i*Kps_i, and step-length is
Kpb_i.The neuron number of input layer is the pixel number of 3 passages of image.
Alternatively, during step S104 is performed, stochastic gradient descent method can be used to carry out the first default convolution
Neural network model is trained, and its learning rate is set to Lr, and momentum term is set to Mo, and the attenuation coefficient of learning rate is set to
Dc；And then detection model training step, above-mentioned training parameter is used, using the image sample data and vehicle of license board information database
Location tags are trained to the first default convolutional neural networks model, until convergence.
Alternatively, perform step S106 during, can using depth learning technology to pre-set image handle model,
Second default convolutional neural networks model is cascaded with the first default convolutional neural networks model for reaching convergence state, is obtained
3rd default convolutional neural networks model.Second default convolutional neural networks model can include 3 sub- network branches.Specifically such as
Shown in lower：
First network branches into Chinese Character Recognition sub-network, comprising 1 input layer, nc_1 convolutional layer, np_1 pond layer,
Nf_1 full articulamentums.Wherein, above-mentioned input layer number is the pixel number of the license plate area image alignd.Every layer
Convolutional layer includes some convolution kernels, and j-th of convolutional layer includes k1_j convolution kernel, and the size of convolution kernel is s1_j*s1_j, step-length
For b1_j.Every layer of pond layer is using maximum pond method, and Chi Huahe size is p1_j*p1_j, and step-length is pb1_j.It is above-mentioned to connect entirely
The neuronal quantity for connecing last layer of output of layer is 32, correspondence above-mentioned 31 chinese characters and 1 question mark character.
Second sub-network branches into Letter identification sub-network, includes 1 input layer, nc_2 convolutional layer, np_2 pond
Layer, nf_2 full articulamentums.Wherein, above-mentioned input layer number is the pixel number of the license plate area image alignd.
Every layer of convolutional layer includes some convolution kernels, and j-th of convolutional layer includes k2_j convolution kernel, and the size of convolution kernel is s2_j*s2_j,
Step-length is b2_j.Every layer of pond layer is using maximum pond method, and Chi Huahe size is p2_j*p2_j, and step-length is pb2_j.It is above-mentioned
The neuronal quantity of last layer of output of full articulamentum is 25, correspondence above-mentioned 24 English alphabet characters and 1 question mark word
Symbol.
3rd sub-network branches into numeral, Letter identification sub-network, and comprising 1 input layer, nc_3 convolutional layer, np_3 are individual
Pond layer, nf_3 full articulamentums and 5 sub- network branches.Wherein, above-mentioned input layer number is the car plate area of alignment
The pixel number of area image.Every layer of convolutional layer includes some convolution kernels, and j-th of convolutional layer includes k3_j convolution kernel, convolution
The size of core is s3_j*s3_j, and step-length is b3_j.Every layer of pond layer is using maximum pond method, and Chi Huahe size is p3_j*
P3_j, step-length is pb3_j.Above-mentioned 5 sub- network branches structures are identical, including z_c convolutional layer, z_c pond layer, z_c individual
Full articulamentum.Wherein, the neuronal quantity of last layer of output of the full articulamentum of each sub-network branch is 35, and correspondence is above-mentioned
24 English alphabet characters, 10 numerical characters, 1 question mark character.
Alternatively, by the first default convolutional neural networks model after training, pre-set image processing model and initialization
Second default convolutional neural networks model is cascaded in order, can obtain the 3rd default convolutional neural networks model.Wherein,
Preliminary car plate detection area image is obtained by the output result of input image data and the first default convolutional neural networks model, made
The input of model is handled for pre-set image, the license plate area image after pre-set image processing model output correction alignment is used as the
The input data of two default convolutional neural networks models.Wherein, the first default convolutional neural networks model is properly termed as car plate again
Convolutional neural networks model is detected, pre-set image processing model is properly termed as car plate alignment model, the second default convolutional Neural again
Network model is properly termed as number-plate number identification convolutional neural networks model again.
Alternatively, during step S108 is performed, stochastic gradient descent method can be used to the 3rd default convolution god
It is trained through network model, wherein, the parameter of the first default convolutional neural networks model is kept freezing, and only training second is preset
The parameter of convolutional neural networks model, its learning rate is set to lr, and momentum term is set to mo, and the attenuation coefficient of learning rate is set
It is set to dc.And then, with above-mentioned training parameter, using the image sample data and vehicle number label pair of license board information database
3rd default convolutional neural networks model is trained, wherein in backpropagation, only updating the 3rd default convolutional neural networks
The parameter of second default convolutional neural networks model in model, until convergence.
Alternatively, can be by new vehicle image data input to the 3rd default volume during step S108 is performed
Product neural network model, new image format can be JPEG, RMP etc., and then, the recognition result of the number-plate number can be obtained.
Alternatively, before step S104 is performed, i.e., pre- to first according to the first license plate area image and positional information
If before convolutional neural networks model is trained, this method can also include：To in the first default convolutional neural networks model
Network parameter carry out initialization process.
Alternatively, Fig. 2 is the flow signal of another optional number-plate number recognition methods according to embodiments of the present invention
Figure, as shown in Fig. 2 step S104, according to the first license plate area image and positional information to the first default convolutional neural networks mould
Type be trained including：
Step S202, is marked, obtains car respectively to the car plate position in the first license plate area image and the number-plate number
Board location tags and number-plate number label；
Step S204, license board information database is set up according to car plate location tags and number-plate number label, wherein, car plate letter
Cease and multiple data samples are included in database；
Step S206, according to the default convolutional neural networks model of data sample training first.
It is alternatively possible to carry out postsearch screening to the first license plate area image, remove picture quality and image content does not conform to
It is required that view data, so as to obtain the first license plate area image of M opening and closing lattice, and then, can be to garbled M the first cars
Board area image adds car plate position and number-plate number label, is finally built according to whole car plate location tags and number-plate number label
Vertical license board information database, license board information database can provide data supporting to the training of each above-mentioned model.
Alternatively, before step S108 is performed, i.e., pre- to the 3rd according to the second license plate area image and number information
If before convolutional neural networks model is trained, this method can also include：Model is handled according to pre-set image pre- to first
If the first license plate area image of convolutional neural networks model output carries out two dimensional affine conversion process, the second license plate area is obtained
Image.
Alternatively, the first license plate area image is inputted to the first default convolutional neural networks model and obtains the inspection of car plate position
Result is surveyed, non-maximize is carried out to car plate detection position frame and is suppressed, the first license plate area image of preliminary treatment can be obtained, entered
And, two dimensional affine conversion is carried out to the first license plate area image of above-mentioned preliminary treatment, the second license plate area image can be obtained.
Alternatively, Fig. 3 is the flow signal of another optional number-plate number recognition methods according to embodiments of the present invention
Figure, as shown in figure 3, this method can include：
Step S302：Image sample data of the collection comprising vehicle and car plate, car plate position, the number-plate number are carried out to image
Label is marked, and obtains license board information database；
Step S303：Build car plate detection convolutional neural networks model D_CNN (the i.e. above-mentioned first default convolutional neural networks
Model), and carry out the initialization of network parameter；
Step S306：Using the image pattern in license board information database and car plate location tags, training car plate detection volume
Product neural network model D_CNN, until convergence；
Step S308：Car plate alignment model (i.e. above-mentioned pre-set image processing model) is built, to car plate detection convolutional Neural
The license plate area that network model is obtained carries out two dimensional affine conversion, the license plate area after being alignd；
Step S310：Build number-plate number identification convolutional neural networks model R_CNN (the i.e. above-mentioned second default convolutional Neurals
Network model), and carry out the initialization of network parameter；
Step S312：By the car plate detection convolutional neural networks model D_CNN trained, car plate alignment model, initialization
Number-plate number identification convolutional neural networks model R_CNN cascaded, obtain Car license recognition totality deep layer network model T_CNN
(the i.e. above-mentioned 3rd default convolutional neural networks model)；
Step S313：Using the image pattern in license board information database and number-plate number label, training Car license recognition is total
Body deep layer network model T_CNN, until convergence；
Step S316：The new view data of input is obtained to the overall deep layer network model T_CNN of Car license recognition trained
The recognition result of the number-plate number.
Alternatively, performing step S302 process can include：
Step S11, the image sample data comprising vehicle and car plate is obtained using camera, Internet resources；
Step S12, postsearch screening is carried out to described image sample data, removal picture quality and image content do not conform to will
The view data asked, obtains the image sample data of M opening and closing lattice.M >=5000.Preferably, M >=20000；
Step S13, to the garbled M image sample data addition car plate positions and number-plate number label, wherein
Every image includes at least one car plate.
Alternatively, the car plate location tags include the position of all car plates in image, and each position is by two coordinates
The upper left corner and the lower right corner of four points, respectively car plate edge frame that constitute；The number-plate number label is comprising all in image
The number of car plate, 7 characters that each number is made up of Chinese character, numeral, letter and question mark, wherein question mark is represented due to mould
The character that the problems such as pasting, block is beyond recognition.
Alternatively, the chinese character includes 31 Chinese characters：Capital, Shanghai, Tianjin, Chongqing, Ji, Shanxi, illiteracy, the Liao Dynasty, Ji, black, Soviet Union, Zhejiang,
It is Anhui, Fujian, Jiangxi, Shandong, Henan, Hubei Province, Hunan, Guangdong, osmanthus, fine jade, river, expensive, cloud, Tibetan, Shan, sweet, blue or green, peaceful, new；The numerical character includes 10
Numeral：0、1、2、3、4、5、6、7、8、9；The alphabetic character includes 24 English alphabets：A、B、C、D、E、F、G、H、J、K、L、
M、N、P、Q、R、S、T、U、V、W、X、Y、Z。
Alternatively, performing step S304 process can include：
Step S21, multilayer convolutional neural networks are built using depth learning technology, including input data layer, Nc
Convolutional layer, Np pond layer, Nf full articulamentums.Wherein, every layer of convolutional layer includes some convolution kernels, and i-th of convolutional layer is included
Ki convolution kernel, the size of convolution kernel is Ks_i*Ks_i, and step-length is Kb_i.Every layer of pond layer is using maximum pond method, Chi Huahe
Size be Kps_i*Kps_i, step-length is Kpb_i.The neuron number of the input layer is the pixel of 3 passages of image
Number.Wherein, the Nc ∈ [5,20], Np ∈ [2,20], Nf ∈ [1,5], Ki ∈ [16,512], Ks_i ∈ [1,9] and for odd number,
Kb_i ∈ [1,5] and Kb_i≤Ks_i, Kps_i ∈ [1,5], Kpb_i ∈ [1,5] and Kpb_i≤Kps_i.
Step S22, random initializtion is carried out to network parameter.
Fig. 4 is the structural representation that according to embodiments of the present invention a kind of optional first presets convolutional neural networks model
Figure, as shown in figure 4, the first default convolutional neural networks model D_CNN includes：
Input data layer, inputs RGB Three Channel Color images, picture size is Width*Height；
First convolutional layer C1, comprising 16 convolution kernels, the size of convolution kernel is 3*3, and step-length is 1, and zero padding parameter is set to
1, using ReLU activation primitives；
First pond layer P1, using maximum pond method, Chi Huahe size is 2*2, and step-length is 2；
Second convolutional layer C2, comprising 32 convolution kernels, the size of convolution kernel is 3*3, and step-length is 1, and zero padding parameter is set to
1, using ReLU activation primitives；
Second pond layer P2, using maximum pond method, Chi Huahe size is 2*2, and step-length is 2；
3rd convolutional layer C3, comprising 64 convolution kernels, the size of convolution kernel is 3*3, and step-length is 1, and zero padding parameter is set to
1, using ReLU activation primitives；；
3rd pond layer P3, using maximum pond method, Chi Huahe size is 2*2, and step-length is 2；
Volume Four lamination C4, comprising 128 convolution kernels, the size of convolution kernel is 3*3, and step-length is 1, and zero padding parameter is set to
1, using ReLU activation primitives；
4th pond layer P4, using maximum pond method, Chi Huahe size is 2*2, and step-length is 2；
5th convolutional layer C5, comprising 256 convolution kernels, the size of convolution kernel is 3*3, and step-length is 1, and zero padding parameter is set to
1, using ReLU activation primitives；
5th pond layer P5, using maximum pond method, Chi Huahe size is 2*2, and step-length is 2；
6th convolutional layer C6, comprising 512 convolution kernels, the size of convolution kernel is 3*3, and step-length is 1, and zero padding parameter is set to
1, using ReLU activation primitives；
7th convolutional layer C7, comprising 256 convolution kernels, the size of convolution kernel is 3*3, and step-length is 1, and zero padding parameter is set to
1, using ReLU activation primitives；
Full articulamentum FC, includes 490 neurons.
Wherein, Width and Height are respectively the width and height of input picture, Width ∈ [50,1680], Height
∈ [50,1050], it is preferable that Width is set to 448, Height and is set to 448.
Alternatively, performing step S306 process can include：
Step S31, carries out car plate detection convolutional neural networks model D_CNN using stochastic gradient descent method and trains, it is learned
Practise speed and be set to Lr, momentum term is set to Mo, and the attenuation coefficient of learning rate is set to Dc.Wherein, Lr is set to 0.01, Mo and set
10 are set to for 0.9, Dc；
Step S32, with described training parameter, using the image sample data and vehicle location mark of license board information database
Label are trained to car plate detection convolutional neural networks model D_CNN, until convergence.
Alternatively, performing step S308 process can include：
Step S41, the image sample data in license board information database is inputted to car plate detection convolutional neural networks mould
Type D_CNN obtains car plate position testing result, and non-maximize is carried out to car plate detection position frame and is suppressed, with reference to input picture sample
Data obtain preliminary license plate area image；
Step S42, two dimensional affine conversion, the license plate area after being alignd are carried out to above-mentioned preliminary license plate area image
Image.
Alternatively, performing step S310 process can include：
Step S51, multilayer convolutional neural networks, including 3 sub- network branches are built using depth learning technology；
Step S52, random initializtion is carried out to network parameter.
Alternatively, the second default convolutional neural networks model R_CNN can include 3 sub- network branches.Wherein, the first son
Network branches include：
First convolutional layer AC1, comprising 16 convolution kernels, the size of convolution kernel is 5*5, and step-length is 1, and zero padding parameter is set to
1, using ReLU activation primitives；
First pond layer AP1, using maximum pond method, Chi Huahe size is 2*2, and step-length is 1；
Second convolutional layer AC2, comprising 32 convolution kernels, the size of convolution kernel is 5*5, and step-length is 1, and zero padding parameter is set to
1, using ReLU activation primitives；
Second pond layer AP2, using maximum pond method, Chi Huahe size is 2*2, and step-length is 1；
3rd convolutional layer AC3, comprising 64 convolution kernels, the size of convolution kernel is 5*5, and step-length is 1, and zero padding parameter is set to
1, using ReLU activation primitives；
3rd pond layer AP3, using maximum pond method, Chi Huahe size is 2*2, and step-length is 1；
Volume Four lamination AC4, comprising 128 convolution kernels, the size of convolution kernel is 3*3, and step-length is 1, and zero padding parameter is set
For 1, using ReLU activation primitives；
4th pond layer AP4, using maximum pond method, Chi Huahe size is 2*2, and step-length is 1；
5th convolutional layer AC5, comprising 128 convolution kernels, the size of convolution kernel is 3*3, and step-length is 1, and zero padding parameter is set
For 1, using ReLU activation primitives；
First full articulamentum AF1, includes 512 neurons；
Second full articulamentum AF2, comprising 32 neurons, using softmax functions, 31 described chinese characters of correspondence
And 1 question mark character.
Alternatively, the second sub-network branches into Letter identification sub-network, includes 1 input layer, nc_2 convolutional layer, np_2
Individual pond layer, nf_2 full articulamentums.As shown in figure 3, the second sub-network branch includes：
First convolutional layer BC1, comprising 16 convolution kernels, the size of convolution kernel is 5*5, and step-length is 1, and zero padding parameter is set to
1, using ReLU activation primitives；
First pond layer BP1, using maximum pond method, Chi Huahe size is 2*2, and step-length is 1；
Second convolutional layer BC2, comprising 32 convolution kernels, the size of convolution kernel is 5*5, and step-length is 1, and zero padding parameter is set to
1, using ReLU activation primitives；
Second pond layer BP2, using maximum pond method, Chi Huahe size is 2*2, and step-length is 1；
3rd convolutional layer BC3, comprising 64 convolution kernels, the size of convolution kernel is 5*5, and step-length is 1, and zero padding parameter is set to
1, using ReLU activation primitives；
First full articulamentum BF1, includes 256 neurons；
Second full articulamentum BF2, comprising 25 neurons, using softmax functions, 24 described English alphabets of correspondence
Character and 1 question mark character.
Alternatively, the 3rd sub-network branches into numeral, Letter identification sub-network, includes 1 input layer, nc_3 convolution
Layer, np_3 pond layer, nf_3 full articulamentums and 5 sub- network branches.As shown in figure 3, the 3rd sub-network branch bag
Include：
First convolutional layer CC1, comprising 16 convolution kernels, the size of convolution kernel is 5*5, and step-length is 1, and zero padding parameter is set to
1, using ReLU activation primitives；
First pond layer CP1, using maximum pond method, Chi Huahe size is 2*2, and step-length is 1；
Second convolutional layer CC2, comprising 32 convolution kernels, the size of convolution kernel is 5*5, and step-length is 1, and zero padding parameter is set to
1, using ReLU activation primitives；
First full articulamentum CF1, comprising 256 neurons, its result is exported to 5 sub- network branches, each sub-network
Branched structure is identical, including：
First convolutional layer ZC1, comprising 32 convolution kernels, the size of convolution kernel is 3*3, and step-length is 1, and zero padding parameter is set to
1, using ReLU activation primitives；
First pond layer ZP1, using maximum pond method, Chi Huahe size is 2*2, and step-length is 1；
Second convolutional layer ZC2, comprising 32 convolution kernels, the size of convolution kernel is 3*3, and step-length is 1, and zero padding parameter is set to
1, using ReLU activation primitives；
Second pond layer ZP2, using maximum pond method, Chi Huahe size is 2*2, and step-length is 1；
First full articulamentum ZF1, includes 256 neurons；
Second full articulamentum ZF2, comprising 35 neurons, using softmax functions, 24 described English alphabets of correspondence
Character, 10 numerical characters, 1 question mark character.
Alternatively, can be by the car plate detection convolutional neural networks of the pre-training during step S312 is performed
Model D_CNN, car plate alignment model, the number-plate number identification convolutional neural networks model R_CNN of initialization carry out level in order
Connection, obtains the overall deep layer network model T_CNN of Car license recognition.Wherein, by input image data and car plate detection convolutional Neural net
Network model D_CNN output result obtains preliminary car plate detection area image, is used as the input of car plate alignment model, car plate alignment
License plate area image after model output correction alignment, deep layer network model R_CNN input data is recognized as the number-plate number.
Alternatively, during step S314 is performed, stochastic gradient descent method can be used overall to Car license recognition deep
Layer network model (the i.e. above-mentioned 3rd default convolutional neural networks model) is trained, wherein, car plate detection convolutional neural networks
The parameter of model keeps freezing, and only trains the parameter of Car license recognition convolutional neural networks model, and its learning rate is set to lr, moves
Quantifier is set to mo, and the attenuation coefficient of learning rate is set to dc.Wherein, lr is set to 0.02, mo and is set to 0.9, dc to be set to 10.Enter
And, can be using the image sample data and vehicle number label of license board information database to the overall deep layer network mould of Car license recognition
Type is trained, wherein in backpropagation, only updating Car license recognition convolutional Neural in the overall deep layer network model of Car license recognition
The parameter of network model, until convergence.
Alternatively, can be total to Car license recognition by new vehicle image data input during step S316 is performed
Body deep layer network model, new image format can be JPEG, RMP etc..And then, by car plate detection convolutional neural networks model and car
The result of board identification convolutional neural networks model is integrated, and can obtain the overall result of number-plate number identification.
In embodiments of the present invention, using obtain by camera acquisition to history vehicle image in the first license plate area
The positional information of image and the mode of number information, and then according to the first license plate area image and positional information to the first default volume
Product neural network model is trained, until the first default convolutional neural networks model reaches convergence state, and then by pre-
If image processing model, the second default convolutional neural networks model and the first default convolutional neural networks mould for reaching convergence state
Type carries out cascade so as to obtain the 3rd default convolutional neural networks model, and then according to the second license plate area image and number information
3rd default convolutional neural networks model is trained up to the 3rd default convolutional neural networks model reaches convergence state, reached
The Current vehicle image collected is identified to convolutional neural networks model is preset according to the reach convergence state the 3rd
So that the purpose of the recognition result of the number-plate number in Current vehicle image is obtained, it is achieved thereby that lifting number-plate number identification
The degree of accuracy and recognition efficiency, avoid the extraneous unfavorable factor such as environment, illumination, fuzzy interference technique effect, and then solve
Car license recognition is present in the prior art recognition accuracy and less efficient technical problem.
Embodiment 2
Other side according to embodiments of the present invention, additionally provides a kind of number-plate number identifying device, as shown in figure 5,
The device includes：Acquiring unit 501, first processing units 503, second processing unit 505, the 3rd processing unit 507, identification are single
Member 509.
Wherein, acquiring unit 501, for obtain by camera acquisition to history vehicle image in the first license plate area
The positional information and number information of image；First processing units 503, for according to the first license plate area image and positional information pair
First default convolutional neural networks model is trained, until the first default convolutional neural networks model reaches convergence state, its
In, the first default convolutional neural networks model is used to detect the car plate position in the first license plate area image；At second
Unit 505 is managed, for model, the second default convolutional neural networks model to be handled pre-set image and the first of convergence state is reached
Default convolutional neural networks model is cascaded, so that the 3rd default convolutional neural networks model is obtained, wherein, at pre-set image
Reason model is used to carry out registration process to the first license plate area image to obtain the second license plate area image, the second default convolution
Neural network model is used to carry out depth recognition to the number-plate number in the second license plate area image；3rd processing unit 507, is used
The 3rd default convolutional neural networks model is trained according to the second license plate area image and number information, until the 3rd pre-
If convolutional neural networks model reaches convergence state；Recognition unit 509, for according to the reach convergence state the 3rd default convolution
The Current vehicle image collected is identified neural network model, obtains the identification of the number-plate number in Current vehicle image
As a result.
Alternatively, the device can also include：Fourth processing unit, in the first default convolutional neural networks model
Network parameter carry out initialization process.
Alternatively, the first processing units 503 can include：Subelement is marked, in the first license plate area image
Car plate position and the number-plate number be marked respectively, obtain car plate location tags and number-plate number label；Subelement is created, is used
In setting up license board information database according to car plate location tags and number-plate number label, wherein, included in license board information database
Multiple data samples；First processing subelement, for according to the default convolutional neural networks model of data sample training first.
Alternatively, the device can also include：5th processing unit, it is pre- to first for handling model according to pre-set image
If the first license plate area image of convolutional neural networks model output carries out two dimensional affine conversion process, the second license plate area is obtained
Image.
Another aspect according to embodiments of the present invention, additionally provides a kind of storage medium, and storage medium includes storage
Program, wherein, equipment performs above-mentioned number-plate number recognition methods where controlling storage medium when program is run.
Another aspect according to embodiments of the present invention, additionally provides a kind of processor, it is characterised in that processor is used for
Operation program, wherein, program performs the above-mentioned number-plate number recognition methods when running.
In embodiments of the present invention, believe the position of the license plate area image in the history vehicle image collected using obtaining
The mode of breath and number information, so as to be entered according to license plate area image and positional information to the first default convolutional neural networks model
Row training restrains up to the first default convolutional neural networks model, and then pre- to second according to license plate area image and number information
If convolutional neural networks model is trained until the second default convolutional neural networks model convergence, has reached default according to second
The number-plate number obtained in Current vehicle image is identified to the Current vehicle image collected in convolutional neural networks model
The purpose of recognition result, it is achieved thereby that the degree of accuracy of lifting number-plate number identification and recognition efficiency, avoiding environment, illumination, mould
The technique effect of the interference of the extraneous unfavorable factor such as paste, and then solve the recognition accuracy that Car license recognition is present in the prior art
With less efficient technical problem.
The embodiments of the present invention are for illustration only, and the quality of embodiment is not represented.
In the above embodiment of the present invention, the description to each embodiment all emphasizes particularly on different fields, and does not have in some embodiment
The part of detailed description, may refer to the associated description of other embodiment.
In several embodiments provided herein, it should be understood that disclosed technology contents, others can be passed through
Mode is realized.Wherein, device embodiment described above is only schematical, such as division of described unit, Ke Yiwei
A kind of division of logic function, can there is other dividing mode when actually realizing, such as multiple units or component can combine or
Person is desirably integrated into another system, or some features can be ignored, or does not perform.Another, shown or discussed is mutual
Between coupling or direct-coupling or communication connection can be the INDIRECT COUPLING or communication link of unit or module by some interfaces
Connect, can be electrical or other forms.
The unit illustrated as separating component can be or may not be it is physically separate, it is aobvious as unit
The part shown can be or may not be physical location, you can with positioned at a place, or can also be distributed to multiple
On unit.Some or all of unit therein can be selected to realize the purpose of this embodiment scheme according to the actual needs.
In addition, each functional unit in each embodiment of the invention can be integrated in a processing unit, can also
That unit is individually physically present, can also two or more units it is integrated in a unit.Above-mentioned integrated list
Member can both be realized in the form of hardware, it would however also be possible to employ the form of SFU software functional unit is realized.
If the integrated unit is realized using in the form of SFU software functional unit and as independent production marketing or used
When, it can be stored in a computer read/write memory medium.Understood based on such, technical scheme is substantially
The part contributed in other words to prior art or all or part of the technical scheme can be in the form of software products
Embody, the computer software product is stored in a storage medium, including some instructions are to cause a computer
Equipment (can for personal computer, server or network equipment etc.) perform each embodiment methods described of the invention whole or
Part steps.And foregoing storage medium includes：USB flash disk, read-only storage (ROM, Read-Only Memory), arbitrary access are deposited
Reservoir (RAM, Random Access Memory), mobile hard disk, magnetic disc or CD etc. are various can be with store program codes
Medium.
Described above is only the preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, under the premise without departing from the principles of the invention, some improvements and modifications can also be made, these improvements and modifications also should
It is considered as protection scope of the present invention.
Claims (10)
1. a kind of number-plate number recognition methods, it is characterised in that including：
Obtain by camera acquisition to history vehicle image in the first license plate area image positional information and number information；
The first default convolutional neural networks model is trained according to the first license plate area image and the positional information,
Until the described first default convolutional neural networks model reaches convergence state, wherein, the described first default convolutional neural networks mould
Type is used to detect the car plate position in the first license plate area image；
Handle pre-set image model, the second default convolutional neural networks model and reach the described first pre- of the convergence state
If convolutional neural networks model is cascaded, so that the 3rd default convolutional neural networks model is obtained, wherein, the pre-set image
Processing model is used to the first license plate area image is carried out registration process to obtain the second license plate area image, and described the
Two default convolutional neural networks models are used to carry out depth recognition to the number-plate number in the second license plate area image；
The described 3rd default convolutional neural networks model is carried out according to the second license plate area image and the number information
Training, until the described 3rd default convolutional neural networks model reaches convergence state；
According to the reach the convergence state the described 3rd default convolutional neural networks model to the Current vehicle image that collects
It is identified, obtains the recognition result of the number-plate number in the Current vehicle image.
2. according to the method described in claim 1, it is characterised in that according to the first license plate area image and the position
Before information is trained to the first default convolutional neural networks model, methods described also includes：To the described first default convolution
Network parameter in neural network model carries out initialization process.
3. according to the method described in claim 1, it is characterised in that described according to the first license plate area image and institute's rheme
Confidence breath the first default convolutional neural networks model is trained including：
The car plate position in the first license plate area image and the number-plate number are marked respectively, car plate is obtained
Location tags and number-plate number label；
License board information database is set up according to the car plate location tags and the number-plate number label, wherein, the car plate letter
Cease and multiple data samples are included in database；
According to the default convolutional neural networks model of data sample training described first.
4. according to the method described in claim 1, it is characterised in that according to the second license plate area image and the number
Before information is trained to the described 3rd default convolutional neural networks model, methods described also includes：
First car plate that model is exported to the described first default convolutional neural networks model is handled according to the pre-set image
Area image carries out two dimensional affine conversion process, obtains the second license plate area image.
5. a kind of number-plate number identifying device, it is characterised in that including：
Acquiring unit, for obtain by camera acquisition to history vehicle image in the first license plate area image position letter
Breath and number information；
First processing units, for presetting convolutional Neural to first according to the first license plate area image and the positional information
Network model is trained, until the described first default convolutional neural networks model reaches convergence state, wherein, described first is pre-
If convolutional neural networks model is used to detect the car plate position in the first license plate area image；
Second processing unit, for handling pre-set image model, the second default convolutional neural networks model and reaching the receipts
The the described first default convolutional neural networks model for holding back state is cascaded, so as to obtain the 3rd default convolutional neural networks mould
Type, wherein, pre-set image processing model is used to carrying out registration process to the first license plate area image to obtain the
Two license plate area images, the described second default convolutional neural networks model is used for the car plate in the second license plate area image
Number carries out depth recognition；
3rd processing unit, for presetting convolution to the described 3rd according to the second license plate area image and the number information
Neural network model is trained, until the described 3rd default convolutional neural networks model reaches convergence state；
Recognition unit, for according to the reach the convergence state the described 3rd default convolutional neural networks model to collecting
Current vehicle image is identified, and obtains the recognition result of the number-plate number in the Current vehicle image.
6. device according to claim 5, it is characterised in that described device also includes：
Fourth processing unit, for being carried out to the network parameter in the described first default convolutional neural networks model at initialization
Reason.
7. device according to claim 5, it is characterised in that the first processing units include：
Subelement is marked, for entering respectively to the car plate position in the first license plate area image and the number-plate number
Line flag, obtains car plate location tags and number-plate number label；
Subelement is created, for setting up license board information database according to the car plate location tags and the number-plate number label,
Wherein, multiple data samples are included in the license board information database；
First processing subelement, for according to the default convolutional neural networks model of data sample training described first.
8. device according to claim 5, it is characterised in that described device also includes：
5th processing unit, it is defeated to the described first default convolutional neural networks model for handling model according to the pre-set image
The the first license plate area image gone out carries out two dimensional affine conversion process, obtains the second license plate area image.
9. a kind of storage medium, it is characterised in that the storage medium includes the program of storage, wherein, in described program operation
When control the storage medium where 1 number-plate number into claim 4 described in any one of equipment perform claim requirement know
Other method.
10. a kind of processor, it is characterised in that the processor is used for operation program, wherein, right of execution when described program is run
Profit requires 1 number-plate number recognition methods into claim 4 described in any one.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710198950.3A CN106980854A (en) | 2017-03-29 | 2017-03-29 | Number-plate number recognition methods, device, storage medium and processor |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710198950.3A CN106980854A (en) | 2017-03-29 | 2017-03-29 | Number-plate number recognition methods, device, storage medium and processor |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106980854A true CN106980854A (en) | 2017-07-25 |
Family
ID=59338555
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710198950.3A Pending CN106980854A (en) | 2017-03-29 | 2017-03-29 | Number-plate number recognition methods, device, storage medium and processor |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106980854A (en) |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107679452A (en) * | 2017-08-28 | 2018-02-09 | 中国电子科技集团公司第二十八研究所 | Goods train license number real-time identifying system based on convolutional neural networks under big data |
| CN107704857A (en) * | 2017-09-25 | 2018-02-16 | 北京邮电大学 | A kind of lightweight licence plate recognition method and device end to end |
| CN107944450A (en) * | 2017-11-16 | 2018-04-20 | 深圳市华尊科技股份有限公司 | A kind of licence plate recognition method and device |
| CN108256516A (en) * | 2017-12-01 | 2018-07-06 | 桂林远望智能通信科技有限公司 | A kind of region licence plate recognition method and system |
| CN108334881A (en) * | 2018-03-12 | 2018-07-27 | 南京云创大数据科技股份有限公司 | A kind of licence plate recognition method based on deep learning |
| CN108491827A (en) * | 2018-04-13 | 2018-09-04 | 腾讯科技（深圳）有限公司 | A kind of vehicle checking method, device and storage medium |
| CN108875746A (en) * | 2018-05-17 | 2018-11-23 | 北京旷视科技有限公司 | A kind of licence plate recognition method, device, system and storage medium |
| CN108898134A (en) * | 2018-06-27 | 2018-11-27 | 北京字节跳动网络技术有限公司 | number identification method, device, terminal device and storage medium |
| CN108921151A (en) * | 2018-05-31 | 2018-11-30 | 四川物联亿达科技有限公司 | A kind of full Vehicle License Plate Recognition System of common camera based on deep learning |
| CN109472262A (en) * | 2018-09-25 | 2019-03-15 | 平安科技（深圳）有限公司 | Licence plate recognition method, device, computer equipment and storage medium |
| CN109685023A (en) * | 2018-12-27 | 2019-04-26 | 深圳开立生物医疗科技股份有限公司 | A kind of facial critical point detection method and relevant apparatus of ultrasound image |
| CN109858309A (en) * | 2017-11-30 | 2019-06-07 | 东软集团股份有限公司 | A kind of method and apparatus identifying Road |
| CN110162454A (en) * | 2018-11-30 | 2019-08-23 | 腾讯科技（深圳）有限公司 | Game running method and device, storage medium and electronic device |
| WO2019174405A1 (en) * | 2018-03-14 | 2019-09-19 | 台达电子工业股份有限公司 | License plate identification method and system thereof |
| CN110414451A (en) * | 2019-07-31 | 2019-11-05 | 深圳市捷顺科技实业股份有限公司 | It is a kind of based on end-to-end licence plate recognition method, device, equipment and storage medium |
| CN110569836A (en) * | 2018-06-06 | 2019-12-13 | 北京深鉴智能科技有限公司 | variable-length character string identification method and device |
| CN110584654A (en) * | 2019-10-09 | 2019-12-20 | 中山大学 | Multi-mode convolutional neural network-based electrocardiosignal classification method |
| CN110738225A (en) * | 2018-07-19 | 2020-01-31 | 杭州海康威视数字技术股份有限公司 | Image recognition method and device |
| CN110781880A (en) * | 2018-07-27 | 2020-02-11 | 业纳交通解决方案英国有限公司 | Method and device for recognizing license plate of vehicle |
| CN111539337A (en) * | 2020-04-26 | 2020-08-14 | 上海眼控科技股份有限公司 | Vehicle posture correction method, device and equipment |
| CN111626578A (en) * | 2020-05-18 | 2020-09-04 | 上海东普信息科技有限公司 | Distribution method, device, equipment and storage medium of logistics truck |
| CN112241640A (en) * | 2019-07-18 | 2021-01-19 | 杭州海康威视数字技术股份有限公司 | Graphic code determination method and device and industrial camera |
| CN114724128A (en) * | 2022-03-21 | 2022-07-08 | 北京卓视智通科技有限责任公司 | License plate recognition method, device, equipment and medium |
| CN116562338A (en) * | 2022-01-27 | 2023-08-08 | 美的集团(上海)有限公司 | Multi-branch convolution structure, neural network model, and determination method and determination device thereof |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105046196A (en) * | 2015-06-11 | 2015-11-11 | 西安电子科技大学 | Front vehicle information structured output method base on concatenated convolutional neural networks |
| US20160086015A1 (en) * | 2007-01-09 | 2016-03-24 | Si Corporation | Method and system for automated face detection and recognition |
| CN106446895A (en) * | 2016-10-28 | 2017-02-22 | 安徽四创电子股份有限公司 | License plate recognition method based on deep convolutional neural network |
-
2017
- 2017-03-29 CN CN201710198950.3A patent/CN106980854A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160086015A1 (en) * | 2007-01-09 | 2016-03-24 | Si Corporation | Method and system for automated face detection and recognition |
| CN105046196A (en) * | 2015-06-11 | 2015-11-11 | 西安电子科技大学 | Front vehicle information structured output method base on concatenated convolutional neural networks |
| CN106446895A (en) * | 2016-10-28 | 2017-02-22 | 安徽四创电子股份有限公司 | License plate recognition method based on deep convolutional neural network |
Non-Patent Citations (2)
| Title |
|---|
| CHRISTIAN GERBER 等: "Number Plate Detection with a Multi-Convolutional Neural Network Approach with Optical Character Recognition for Mobile Devices", 《JOURNAL OF INFORMATION PROCESSING SYSTEMS》 * |
| HUI LI 等: "Reading Car License Plates Using Deep Convolutional Neural Networks and LSTMs", 《ARXIV:1601.05610V1 [CS.CV]》 * |
Cited By (37)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107679452A (en) * | 2017-08-28 | 2018-02-09 | 中国电子科技集团公司第二十八研究所 | Goods train license number real-time identifying system based on convolutional neural networks under big data |
| CN107704857A (en) * | 2017-09-25 | 2018-02-16 | 北京邮电大学 | A kind of lightweight licence plate recognition method and device end to end |
| US10755120B2 (en) | 2017-09-25 | 2020-08-25 | Beijing University Of Posts And Telecommunications | End-to-end lightweight method and apparatus for license plate recognition |
| CN107704857B (en) * | 2017-09-25 | 2020-07-24 | 北京邮电大学 | End-to-end lightweight license plate recognition method and device |
| CN107944450A (en) * | 2017-11-16 | 2018-04-20 | 深圳市华尊科技股份有限公司 | A kind of licence plate recognition method and device |
| CN109858309A (en) * | 2017-11-30 | 2019-06-07 | 东软集团股份有限公司 | A kind of method and apparatus identifying Road |
| CN109858309B (en) * | 2017-11-30 | 2021-04-20 | 东软睿驰汽车技术（上海）有限公司 | Method and device for identifying road route |
| CN108256516A (en) * | 2017-12-01 | 2018-07-06 | 桂林远望智能通信科技有限公司 | A kind of region licence plate recognition method and system |
| CN108334881A (en) * | 2018-03-12 | 2018-07-27 | 南京云创大数据科技股份有限公司 | A kind of licence plate recognition method based on deep learning |
| CN108334881B (en) * | 2018-03-12 | 2022-04-29 | 南京云创大数据科技股份有限公司 | License plate recognition method based on deep learning |
| US11443535B2 (en) | 2018-03-14 | 2022-09-13 | Delta Electronics, Inc. | License plate identification method and system thereof |
| WO2019174405A1 (en) * | 2018-03-14 | 2019-09-19 | 台达电子工业股份有限公司 | License plate identification method and system thereof |
| CN110276342A (en) * | 2018-03-14 | 2019-09-24 | 台达电子工业股份有限公司 | Number plate recognition method and its system |
| CN108491827A (en) * | 2018-04-13 | 2018-09-04 | 腾讯科技（深圳）有限公司 | A kind of vehicle checking method, device and storage medium |
| CN108491827B (en) * | 2018-04-13 | 2020-04-10 | 腾讯科技（深圳）有限公司 | Vehicle detection method and device and storage medium |
| CN108875746A (en) * | 2018-05-17 | 2018-11-23 | 北京旷视科技有限公司 | A kind of licence plate recognition method, device, system and storage medium |
| CN108921151B (en) * | 2018-05-31 | 2022-07-12 | 四川物联亿达科技有限公司 | General camera whole license plate recognition system based on deep learning |
| CN108921151A (en) * | 2018-05-31 | 2018-11-30 | 四川物联亿达科技有限公司 | A kind of full Vehicle License Plate Recognition System of common camera based on deep learning |
| CN110569836B (en) * | 2018-06-06 | 2022-07-12 | 赛灵思电子科技(北京)有限公司 | Variable-length character string identification method and device |
| CN110569836A (en) * | 2018-06-06 | 2019-12-13 | 北京深鉴智能科技有限公司 | variable-length character string identification method and device |
| CN108898134A (en) * | 2018-06-27 | 2018-11-27 | 北京字节跳动网络技术有限公司 | number identification method, device, terminal device and storage medium |
| CN108898134B (en) * | 2018-06-27 | 2020-11-06 | 北京字节跳动网络技术有限公司 | Number identification method and device, terminal equipment and storage medium |
| CN110738225A (en) * | 2018-07-19 | 2020-01-31 | 杭州海康威视数字技术股份有限公司 | Image recognition method and device |
| CN110781880A (en) * | 2018-07-27 | 2020-02-11 | 业纳交通解决方案英国有限公司 | Method and device for recognizing license plate of vehicle |
| CN109472262A (en) * | 2018-09-25 | 2019-03-15 | 平安科技（深圳）有限公司 | Licence plate recognition method, device, computer equipment and storage medium |
| CN110162454B (en) * | 2018-11-30 | 2022-02-08 | 腾讯科技（深圳）有限公司 | Game running method and device, storage medium and electronic device |
| CN110162454A (en) * | 2018-11-30 | 2019-08-23 | 腾讯科技（深圳）有限公司 | Game running method and device, storage medium and electronic device |
| CN109685023A (en) * | 2018-12-27 | 2019-04-26 | 深圳开立生物医疗科技股份有限公司 | A kind of facial critical point detection method and relevant apparatus of ultrasound image |
| CN112241640A (en) * | 2019-07-18 | 2021-01-19 | 杭州海康威视数字技术股份有限公司 | Graphic code determination method and device and industrial camera |
| CN110414451A (en) * | 2019-07-31 | 2019-11-05 | 深圳市捷顺科技实业股份有限公司 | It is a kind of based on end-to-end licence plate recognition method, device, equipment and storage medium |
| CN110414451B (en) * | 2019-07-31 | 2023-11-10 | 深圳市捷顺科技实业股份有限公司 | License plate recognition method, device, equipment and storage medium based on end-to-end |
| CN110584654A (en) * | 2019-10-09 | 2019-12-20 | 中山大学 | Multi-mode convolutional neural network-based electrocardiosignal classification method |
| CN111539337A (en) * | 2020-04-26 | 2020-08-14 | 上海眼控科技股份有限公司 | Vehicle posture correction method, device and equipment |
| CN111626578A (en) * | 2020-05-18 | 2020-09-04 | 上海东普信息科技有限公司 | Distribution method, device, equipment and storage medium of logistics truck |
| CN116562338A (en) * | 2022-01-27 | 2023-08-08 | 美的集团(上海)有限公司 | Multi-branch convolution structure, neural network model, and determination method and determination device thereof |
| CN114724128A (en) * | 2022-03-21 | 2022-07-08 | 北京卓视智通科技有限责任公司 | License plate recognition method, device, equipment and medium |
| CN114724128B (en) * | 2022-03-21 | 2023-10-10 | 北京卓视智通科技有限责任公司 | License plate recognition method, device, equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106980854A (en) | Number-plate number recognition methods, device, storage medium and processor | |
| CN111640101B (en) | Ghost convolution characteristic fusion neural network-based real-time traffic flow detection system and method | |
| CN106570477B (en) | Vehicle cab recognition model building method and model recognizing method based on deep learning | |
| CN109214441A (en) | A kind of fine granularity model recognition system and method | |
| CN107886133A (en) | A kind of underground piping defect inspection method based on deep learning | |
| CN107273872A (en) | The depth discrimination net model methodology recognized again for pedestrian in image or video | |
| CN110378332A (en) | A kind of container terminal case number (CN) and Train number recognition method and system | |
| CN107563396A (en) | The construction method of protection screen intelligent identifying system in a kind of electric inspection process | |
| CN108288035A (en) | The human motion recognition method of multichannel image Fusion Features based on deep learning | |
| CN108334881B (en) | License plate recognition method based on deep learning | |
| CN108280488B (en) | Grippable object identification method based on shared neural network | |
| CN109255284B (en) | Motion trajectory-based behavior identification method of 3D convolutional neural network | |
| CN106778705A (en) | A kind of pedestrian's individuality dividing method and device | |
| CN111832573B (en) | Image emotion classification method based on class activation mapping and visual saliency | |
| CN107292933B (en) | Vehicle color identification method based on BP neural network | |
| CN106709528A (en) | Method and device of vehicle reidentification based on multiple objective function deep learning | |
| CN111507275B (en) | Video data time sequence information extraction method and device based on deep learning | |
| CN112270681B (en) | Method and system for detecting and counting yellow plate pests deeply | |
| CN109145964B (en) | Method and system for realizing image color clustering | |
| CN109344842A (en) | A kind of pedestrian's recognition methods again based on semantic region expression | |
| CN106934355A (en) | In-car hand detection method based on depth convolutional neural networks | |
| CN111967313A (en) | Unmanned aerial vehicle image annotation method assisted by deep learning target detection algorithm | |
| CN105930792A (en) | Human action classification method based on video local feature dictionary | |
| CN107545281B (en) | Single harmful gas infrared image classification and identification method based on deep learning | |
| CN112446417B (en) | Spindle-shaped fruit image segmentation method and system based on multilayer superpixel segmentation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170725 |
|
| RJ01 | Rejection of invention patent application after publication |